US5703959A - Method and device for analyzing particles - Google Patents

Method and device for analyzing particles Download PDF

Info

Publication number
US5703959A
US5703959A US08368053 US36805395A US5703959A US 5703959 A US5703959 A US 5703959A US 08368053 US08368053 US 08368053 US 36805395 A US36805395 A US 36805395A US 5703959 A US5703959 A US 5703959A
Authority
US
Grant status
Grant
Patent type
Prior art keywords
data
gravity
center
measured
corpuscles
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
US08368053
Inventor
Kaoru Asano
Kimiyo Kubo
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sysmex Corp
Original Assignee
Sysmex Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Grant date

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • G01N15/1031Investigating individual particles by measuring electrical or magnetic effects thereof, e.g. onconductivity or capacity
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • G01N15/12Coulter-counters
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • G01N15/14Electro-optical investigation, e.g. flow cytometers
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • G01N2015/1037Associating coulter-counter and optical flow cytometer [OFC]
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
    • G01N15/10Investigating individual particles
    • G01N2015/1062Investigating individual particles counting the particles by other than electro-optical means

Abstract

When a mixture of different kinds of particles, such as blood, is passed through a particle detector such as a flow cytometer to measure various characteristics of each particle and the measurements of two characteristics are plotted oh a two-dimensional rectangular co-ordinate, for example, the resultant dots tend to part into groups (clusters) corresponding to the kinds of the particles. This invention relates to a method and a device for presuming informations of the position of center of gravity, variances, number of dots and likes of each cluster by learning and, based upon these informations, presuming a specific cluster or category to which each measured particle should belong from the corresponding measurements of the particle, and it intends to execute such information processing by a fuzzy clustering technique using a neural network.

Description

This application is a continuation of application Ser. No. 08/076,870, filed Jun. 15, 1993, now abandoned.

BACKGROUND OF THE INVENTION

This invention relates to a method and a device for analyzing plural kinds of particles intermixed with each other, by presuming the kind of each particle from measurements of specific characteristics thereof and counting the number of particles belonging to each kind and, especially, to a system for presuming the kind of each particle as detected based upon distribution data obtained by using a neural network and previously applying informations of the respective characteristics of a number of particles thereto to cause it to learn their states of distribution.

While the above-mentioned particle analyzing device is useful, for example, as a blood-corpuscle analyzer for classifying and counting blood-corpuscles in blood, such device is disclosed, for example, in the Japanese opened patent gazette No. H3(91)-131756. In this prior art device, predetermined characteristic values (hereinunder generalized as "characteristic parameters") of blood-corpuscles belonging to plural kinds (hereinunder generalized as "categories") are applied to an input layer of a neural network of error back-propagation type to cause it to learn supervisedly their states of distribution and, thereafter, the characteristic parameters of each particle as detected are supplied to the input layer to cause it to presume the group (hereinunder referred to as "cluster") or the category to which the particle is to belong, that is, to effect a clustering operation, based upon the resultant data of learning.

However, this device has such a disadvantage in that it may erroneously classify a particle of an unlearned category into a learned category when the characteristic parameters of such particle of unlearned category are applied and, moreover, it needs a long time for learning when the distribution of characteristic parameters are complicated, since it uses a neural network of error back-propagation type.

Accordingly, an object of this invention is to provide a method and a device for analyzing particles which can effect an accurate clustering operation within a short time regardless of abnormal or complicated distribution of the characteristic parameters.

SUMMARY OF THE INVENTION

In order to achieve the above-mentioned object, the method and device of this invention are arranged to apply a time-sequence of measured data consisting of plural characteristic parameters to a neural network at the same time as converting them into distribution data in a characteristic parameter space, and presume cluster informations such as center of gravity of the cluster of each category from the distribution data for clustering the detected particle based upon the presumed cluster informations.

More particularly, a mixture of a number of particles belonging to plural categories is supplied to a particle detecting device such as flow cytometer to successively measure predetermined characteristic parameters of each particle and the measured values are converted into distribution data of a predetermined characteristic parameter space. At the same time, the measured values are sequentially supplied to a neural network having an input layer which includes at least neurons respectively corresponding to the above-mentioned characteristic parameters and an output layer which includes at least neurons respectively corresponding to expected categories to execute unsupervised learning vector quantization for presuming cluster informations including center-of-gravity vector informations of the clusters respectively corresponding to the above-mentioned categories and, then, the above-mentioned distribution data are clustered in accordance with the presumed cluster informations.

The presumed cluster informations may include informations of the number and/or variance of the particles in addition to the above-mentioned center-of-gravity vector informations.

These and other features and functions of this invention will be described in more detail below about a preferred embodiment thereof with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

In the drawings:

FIG. 1 is a block diagram showing the configuration of an embodiment of the particle analyzing device according to this invention;

FIGS. 2a and 2b are explanatory diagrams of a blood-corpuscle detecting method in a blood-corpuscle detecting device in the embodiment of FIG. 1;

FIG. 3 is a block diagram showing internal configuration of an analyzing device in the embodiment of FIG. 1;

FIG. 4 is a diagram showing an example of distribution data formed by data converting means in FIG. 3;

FIG. 5 is an explanatory diagram of an example of the configuration of a neural network in FIG. 3 and a function thereof;

FIG. 6 is a flow chart showing center-of-gravity calculating and particle counting operations in clustering means in FIG. 3;

FIG. 7 is an explanatory diagram of domain division in the operations of FIG. 6;

FIG. 8 is an explanatory diagram of measurement of the distance from a detected particle to the center of gravity of each particle cluster in the operations of FIG. 6;

FIGS. 9a and 9b are diagrams showing an example of distribution data of normal blood-corpuscles and a result of clustering thereof;

FIGS. 10a and 10b are diagrams showing an example of distribution data of abnormal blood-corpuscles and a result of clustering thereof; and

FIGS. 11a and 11b are diagrams showing another example of distribution data of abnormal blood-corpuscles and a results of clustering thereof.

Thoughout the drawings, same reference numerals are given to corresponding constitutional elements.

DESCRIPTION OF PREFERRED EMBODIMENT

FIG. 1 shows an overall configuration of a blood-corpuscle classifying and counting device as a preferred embodiment of the particle analyzing device of this invention. As shown, the device includes a blood-corpuscle detector 2, a signal amplifier 4, an analog-to-digital (A/D) convertor 6, an analyzing device 8 and a display device 10. The blood-corpuscle detector 2 is a well-known device such as flow cytometer and, when a blood specimen including leukocytes only which are separated by preliminary treatment such as dilution and addition of hemolyzing agent, for example, is supplied thereto, it detects blood-corpuscles one by one and sequentially produces electric signals respectively indicative of predetermined characteristics of the respective blood-corpuscles. For instance, when the blood specimen is passed through a fine passageway which can pass the blood-corpuscles one by one and a d.c. current and a high frequency a.c. current are flowed across the passageway at the same time, the d.c. current results in a signal proportional to the size of the cytoplasm of each blood-corpuscle since the d.c. current is shut out by the cytoplasm 12 as shown in FIG. 2a. in contrast, the high frequency a.c. current penetrates the cytoplasm 12 of low density and low impedance and is shut out by only a nucleus or granule 14 of high density and high impedance as shown in FIG. 2b and, therefore, it results in a signal relating to the density and size of the nucleus or granule. Thus, measured values of two kinds of characteristic parameter are obtained from each blood-corpuscle.

These two kinds of output signal of the blood-corpuscle detector 2 are amplified by the amplifier 4 of well-known type and then converted into digital signals by the A/D convertor 6. In this embodiment, both digital signals are respectively quantized into 256 channels in total from channel No. 0 to No. 255. These two kinds of quantized signals obtained from the respective blood-corpuscles will be hereinunder referred to as data X(x1, x2) consisting of a characteristic parameter x1 indicative of the above-mentioned d.c. component and a characteristic parameter x2 indicative of the above-mentioned a.c. component. The data X is then supplied to the analyzing device 8 which is essential to the invention. The analyzing device 8 is composed of a personal computer, microcomputer or microprocessor and functionally includes data processing means 16, a neural network 18 and clustering means 20 as shown in FIG. 3.

The data-X are supplied successively to the data converting means 16 and neural network 18. The data X are converted by the data converting means 16 into distribution data F(x1, x2) in a two-dimensional space as shown in FIG. 4 having the characteristic parameters x1 and x2 as two variables thereof. Since the characteristic parameters x1 and x2 are quantized respectively into 256 channels as described above, this space of distribution consists of 256×256 basic elements in total, each of which stores the number of blood-corpuscles having their characteristic parameters x1 and x2 belonging to two corresponding channels. For example, FIG. 4 shows that the number of blood-corpuscles whose signal has a d.c. component x1 belonging to channel No. 1 and an a.c. component x2 belonging to channel No. 2 is six and, in other words, F(1, 2)=6. Since the leukocytes include various blood-corpuscles such as lymphocytes, monocytes, granulocytes and the like having different values of the characteristic parameters, the distribution data of FIG. 4 tend to form clusters of the respective kinds of blood-corpuscles. The neural network 18 serves to presume the center of gravity, variances and the number of corpuscles of each cluster based upon the input data X.

As shown in FIG. 5, the neural network 18 includes an input layer having two neurons 22 and 24 respectively corresponding to the above-mentioned two characteristic parameter inputs x1 and x2 and an output layer having five neurons 26, 28, 30, 32 and 34 respectively corresponding to expected five classification categories, namely, those of lymphocytes, monocytes, granulocytes, ghost corpuscles (unhemolized blood-platelets and erythrocytes) and other corpuscles (immature blood-corpuscles and likes). The neuron 22 in the input layer for the characteristic parameter x1 is connected to the neurons 26, 28, 30, 32 and 34 of the output layer with weights W11, W12, W13, W14 and W15, respectively, and the input layer neuron 24 for the characteristic parameter x2 is connected to the output layer neurons 26, 28, 30, 32 and 34 with weights W21, W22, W23, W24 and W25, respectively.

If arbitrary values are initially given to the weights W11, W12, W13, W14, W15, W21, W22, W23, W24 and W25 and part or all of the data X to the input layer neurons 22 and 24, the neural network 18 commences a learning operation based upon a method of unsupervised vector quantization. The method of vector quantization is described in detail, for example, in "Neural Networks for Signal Processing", published by Prentice-Hall International Editions.

While the neural network 18 learns a probability density function of the input data X, the above-mentioned weights W11 to W15 and W21 to W25 are updated gradually and finally obtained weight vectors V1 (weights W11 and W21), V2 (weights W12 and W22), V3 (weights W13 and W23), V4 (weights W14 and W24) and V5 (weights W15 and W25) indicate presumed positions of the centers of gravity of the clusters of lymphocytes, monocytes, granulocytes, ghost corpuscles and other corpuscles in the distribution data F(x1, x2), respectively. In the case of normal specimen, the weight vector V5 may be left unchanged from its initial value.

Upon completion of learning, the neural network 18 can presume the number of blood-corpuscles of each kind from relatively few input data since it has learned the probability density function of the input data. More specifically, counters respectively corresponding to the lymphocytes, monocytes, granulocytes, ghost corpuscles and other corpuscles are provided and their counts are assumed as m1, m2, m3, m4 and m5. When a certain datum X1 consists of d.c. and a.c. components x11 and x12, the values of (x11 -W11)2 +(x21 -W21)2, (x11 -W12)2 +(x21 -W22)2, (x11 -W13)2 +(x21 -W23)2, (x11 -W14)2 +(x21 -W24)2 and (x11 -W15)2 +(x21 -W25)2 are calculated respectively. In other words, the distances between signal and weight vectors dn =∥x1 -Wn ∥ are sought where n=1, . . . 5 and Wn =(W1n, W2n). The least one of these distances is selected. For example, when the distance d1 or (x11 -W11)2 +(x21 -W21)2 is the least, the count of the corresponding lymphocyte counter is incremented by one. Similar calculations are executed on the other data supplied sequentially thereafter and the least distances are selected to increment the corresponding counters one by one. Then, the finally obtained counts m1 to m5 of the respective counters give the ratio of the presumed numbers of blood-corpuscles belonging to the respective categories, if the respective clusters of the blood-corpuscles are same in distribution. In the case of normal specimen, m5 may be absent and, therefore, the corresponding cluster is not assumed to exist when m5 is less than a predetermined value.

When the numbers of blood-corpuscles are presumed, the data belonging to the respective categories are stored and the variances of the corpuscles in the respective clusters are calculated on the basis thereof. For example, when X1 (x11, x21), X3 (x13, x23) and X7 (x17, x27) are stored as the data belonging to the lymphocyte category, the presumed variance s1 of the lymphocytes is calculated by dividing the sum of total square deviations of x11, x13 and x17 from their mean value and total square deviations of x21, x23 and x27 from their mean value by the number of corpuscles. Presumed variances s2, s3, s4 and s5 of the other categories are similarly obtained. In this case also, s5 need not be calculated if the specimen is normal.

The blood classifying and counting operation might be completed with the above presumed numbers of corpuscles mn only if the corpuscles belonging to the respective categories have exhibited the same distribution, the presumed numbers mn and variances sn (n=1, . . . 5) as obtained above are of approximate values since their distributions are not always same in practice. Therefore, the clustering means 20 clusters the distribution data F(x1, x2) obtained from the data converting means 16 by using these approximate values and the presumed centers of gravity Vn (n=1, . . . 5) of the respective clusters, which were calculated previously. The operation of the clustering means 20 will be described below with reference to the flow chart of FIG. 6.

Initial division of the distribution data F(x1, x2) is carried out first, after a memory value K indicating the number of iterations is set to one (step S1). In this initial division, the distribution of the basic elements in the distribution data of FIG. 4 is divided into four fixed domains A1, A2, A3 and A4 in which the lymphocytes, monocytes, granulocytes and ghost corpuscles are probably distributed respectively, as shown in FIG. 7, regardless of the number of corpuscles stored in each element. In this case, no domain is established for the other corpuscles. Denoting the clusters of lymphocytes, monocytes, granulocytes, ghost corpuscles and other corpuscles with C1, C2, C3, C4 and C5, respectively, the membership values of the blood-corpuscles existing in the above fired domains A1, A2, A3 and A4 to the respective clusters C1, C2, C3 and C4 are assumed as all one (1). The blood-corpuscles belonging to the cluster C5 are undetermined. The domain A0 in FIG. 7 is a domain having undetermined memberships to all clusters C1 to C5.

The initial centers of gravity are determined next (step S2). In this case, assuming the other corpuscles to exist also and using the center-of-gravity vectors presumed by the neural network 18, the initial center-of-gravity positions of the respective clusters C1, C2, C3, C4 and C5 are determined at (xG11, xG21), (xG12, xG22), (xG13, xG23), (xG14, xG24) and (xG15, xG25), respectively.

Next, membership values of the corpuscles which are not included in any of the fixed domains A1, A2, A3 and A4, that is, the corpuscles distributing in the domain A0 of FIG. 7, to the clusters C1, C2, C3, C4 and C5 are calculated respectively (step S3). To this end, the distance L from each blood-corpuscle in the domain A0 to the center of gravity of each cluster is calculated first. This distance is not an Euclidean distance and it is defined as the minor radius of an ellipse 38 having its center positioned at the center of gravity (xG1n, xG2n) and its major axis inclined by an angle θn with respect to the X1 -axis of FIG. 7 and passing the corpuscle 36 in question, as shown in FIG. 8 where the suffix n is omitted for simplicity. The angle θn has a value which is previously selected experientially for each cluster Cn. The reason why the distance L is so defined is that the actual distributions of all clusters are substantially elliptic in shape. Thus, all corpuscles lying on the same ellipse should be same in distance from the center of gravity and would have the same membership to the cluster Cn as described below.

The distance L can be calculated as follows. If new co-ordinate axes X1 " and X2 " are established along the major and minor axes of the ellipse 38 and the minor and major radii of the ellipse are expressed with L and kL (k is a proportional constant) respectively the point 36 (x1 ", X2 ") on the ellipse is given by the following equation. ##EQU1## It is solved as follows as looking for the minor radius L. ##EQU2## As is understood from FIG. 8, the relation between the co-ordinate (x1 ", x2 ") and the original co-ordinate (x1, x2) of the point 36 is given by the following equations.

X.sub.1 =cosθ.sub.n (x.sub.1 -x.sub.G1n)+sinθ.sub.n (x.sub.2 -x.sub.G2n)                                               (3)

X.sub.2 =-sinθ.sub.n (x.sub.1 -x.sub.G1n)+cosθ.sub.n (x.sub.2 -x.sub.G2n)                                               (4)

By applying the equations 3 and 4 to the equation 2 and using the inclination angle θn of each cluster, the distance Ln(36) from the blood-corpuscle 36 to the center of gravity of each cluster Cn is calculated. Then, using the presumed number and variance mn and sn of the corpuscles in each cluster Cn which are obtained from the neural network 18 as described above the membership Un(36) of the corpuscle 36 to the cluster Cn is given by the following equation. ##EQU3## The above equations are based upon such an idea in that the membership is not only inversely proportional to the square distance to the center of gravity of each cluster, but also proportional to the number of corpuscles and the standard deviation of the cluster.

The memberships of each blood-corpuscle in the domain A0 to the respective clusters Cn are calculated as above and each blood-corpuscle is deemed to belong to the respective clusters Cn at rates or weights corresponding to its memberships Un. The positions of the centers of gravity of the respective clusters are corrected taking such weights of each blood-corpuscle with respect-thereto into consideration. In other words, the weighted centers of gravity are calculated (step S4). The co-ordinates xG1n ' and xG2n ' of the weighted center of gravity of the cluster Cn are given by the following equations. ##EQU4## where i and j are quantized channel numbers of the characteristic parameters X1 and X2 of the distribution data F(x1, x2) of FIG. 4 and both of them take such values as 0, 1, 2, . . . 255 in this embodiment. Unij is a membership of the basic element belonging to channel Nos. i and j and, Nij is the number of blood-corpuscles in this element.

After calculating the weighted center of gravity of each cluster as above, it is compared with the corresponding initial center of gravity to judge whether the difference D therebetween is equal to or less than a predetermined value D0 or not (step S5). The value of D0 may be zero, for example. If D is not greater than D0 in any cluster, the distances from each blood-corpuscle to respective clusters are not ascertained yet and it is deemed that no membership of this blood-corpuscle to that cluster is decided clearly. In this case, the initial center of gravity of each cluster is updated with the calculated weighted center of gravity and it is judged whether the stored number of iterations K is equal to a predetermined value K0 or not (step S6). If not, the value of K is incremented by one (step S7) and step S3 is resumed for repeating the same operation as updating the center of gravity. If K=K0 is achieved during such repetition, such words as "ANALYSIS IMPOSSIBLE" are displayed by the display device 10 to finish the processing (step S8). On the other hand, if D≦D0 is achieved in step S5, the memberships of each blood-corpuscle to all clusters have been ascertained and, therefore, the number of the blood-corpuscles in each cluster is calculated based thereupon (step S9). This calculation is executed by either one of two methods as the following.

In the first method, the cluster corresponding to the greatest membership of each blood-corpuscle in the domain A0 is designated as a cluster to which that blood-corpuscle belongs. For example, when a certain basic element contains ten blood-corpuscles and the memberships U1, U2, U3, U4 and U5 of these corpuscles to the clusters C1, C2, C3, C4 and C5 of lymphocytes, monocytes, granulocytes, ghost corpuscles and other corpuscles are 0.95, 0.03, 0.02, 0.00 and 0.00, respectively, all of these ten corpuscles are designated as belonging to the lymphocyte cluster C1. This method is suitable in such a case in that a blood-corpuscle in question exhibits an especially large membership to specific one of the clusters and the clusters are clearly separated. On the other hand, in the second method, the blood-corpuscles in each basic element are shared to the respective clusters at the rate of memberships. Fox example, when the basic element in question contains ten blood-corpuscles and the memberships U1, U2, U3, U4 and U5 of these corpuscles are 0.2, 0.5, 0.3, 0.0 and 0.0, respectively, the memberships other than U4 and U5 are close to each other and it appears that the clusters C1, C2 and C3 are partially overlapping. If all ten corpuscles are shared only to the monocyte cluster of the greatest membership in accordance with the first method in this case, a large counting error may be caused. Therefore, in accordance with proportional distribution based upon the memberships, two, five, three, zero and zero corpuscles are shared respectively to the clusters C1, C2, C3, C4 and C5. This method is effective when the distribution patterns of the respective clusters overlap partially as above. Thus, the clusters to which each blood-corpuscle belongs are decided and the number of corpuscles of each cluster is calculated.

Since in this embodiment, the final center of gravity of each cluster and the major and minor radii of the ellipse representing spread of its distribution are calculated as described above, it is possible to compare these values with their normal values to diagnose their deviation from the normal values, that is, a state of health.

Next, the description will be made about a result of actual execution of this embodiment using a commercially available thirty-two bit personal computer as the analyzing device 8. FIGS. 9a and 9b show scattergrams of a normal specimen and FIGS. 10a, 10b, 11a and 11b show scattergrams of two abnormal specimens. FIGS. 9a, 10a and 11a show their distribution data F(x1, x2) before clustering and FIGS. 9b, 10b and 11b show those after clustering. Both normal and abnormal specimens are correctly clustered in this embodiment.

TABLE 1 shows the centers of gravity presumed by the neural network 18 and the centers of gravity finally decided by the succeeding clustering operation, based upon the distribution data F(x1, x2) of specimen Nos. 1, 2 and 3 of FIGS. 9a, 10a and 11a. Both are substantially approximate to each other and utility of presumption using the neural network 18 is understood therefrom.

TABLE 2 shows the variances presumed by the neural network 18 and the variances finally decided by the clustering operation, based upon the same distribution data. Although some difference appears therebetween, the presumption has come nearly true.

TABLE 3 shows the numbers of times of operation needed for calculation of the final center of gravity through the clustering operation of the data of the above specimen Nos. 1, 2 and 3, when the presumption by the neural network 18 is used and when it is not used. Although there is no difference between the two in the case of normal specimen No. 1, the neural network shows much less values in the case of abnormal specimen Nos. 2 and 3 and, especially, no good result is obtained without use of the neural network in the case of abnormal specimen No. 3.

                                  TABLE 1__________________________________________________________________________(Center of gravity)   Ghost C4           Lympho C1                   Mono C2  Granulo C3                                      Others C5Specimen   P   F   P   F   P    F   P    F    P   F__________________________________________________________________________No.1(Normal)   22, 25       23, 25           55, 34               54, 31                   96, 66                        101, 70                            165, 79                                 166, 80                                      --  --No.2(Abnormal)   19, 20       18, 20           56, 36               50, 32                   122, 93                        128, 95                            198, 96                                  203, 101                                      --  --No.3(Abnormal)   20, 21       20, 21           50, 33               51, 31                   92, 63                         93, 64                             157, 159                                 159, 79                                      183, 61                                          186, 70__________________________________________________________________________ (P: Presumed; F: Final)

                                  TABLE 2__________________________________________________________________________(Variance)   Ghost C4        Lympho C1              Mono C2                    Granulo C3                            Others C5Specimen   P  F P  F  P  F  P   F   P  F__________________________________________________________________________No.1(Normal)   68 66        124           123              432                  425                    696 693 -- --No.2(Abnormal)   38 43        134           127              540                 1029                    1050                        1137                            -- --No.3(Abnormal)   61 72        183           152              361                 3320                    705 788 455                               569__________________________________________________________________________ (P: Presumed; F: Final)

              TABLE 3______________________________________(Number of operations)Specimen       Presumed Not presumed______________________________________No. 1 (Normal) 2 times  2 timesNo. 2 (Abnormal)          2 times  4 timesNo. 3 (Abnormal)          2 times  4 times______________________________________

The above embodiment is presented for illustrative purpose only and is not intended as any limitation of the invention. It should be obvious to those skilled in the art that various modifications and variations can be made on this embodiment without departing from the spirit and scope of this invention as defined in the appended claims. For example, the particles to be subjected to the inventive analysis need not always be blood-corpuscles, but may be any other kind of particles such as powdered material. In some of such other particles, the above-mentioned distance to the center of gravity may be an Euclidean distance. Moreover, instead of the above-mentioned d.c. and a.c. impedances as the characteristic parameters of the particles, other characteristic signals such as scattered light and fluorescent intensities may be used. Using three or more characteristic parameters, a data distribution in a space of three or higher dimension may be handled. While the center of gravity, the number of particles and the variances of each cluster are used as the object of presumption by the neural network, the center of gravity may be used alone or together with the number of particles. Furthermore, the configuration of the neural network may be modified arbitrarily, which the number of categories is large, for example a suppressive coupling may be inserted between the neurons of the output layer so that, when a specific neuron is excited, the neighboring neurons are suppressed in response thereto.

Claims (10)

We claim:
1. In an analyzing device, a method of analyzing a mixture of particles belonging to a plurality of categories to count a number of particles belonging to each of the categories, each particle being individually characterized by a corresponding measured data vector, each measured data vector including a plurality of characteristic parameters, each characteristic parameter being representative of a physical characteristic of a corresponding particle, the method comprising steps of:
converting a plurality of measured data vectors representative of a specimen of particles into a spatial distribution of the plurality of measured data vectors based on spatial dimensions corresponding to the plurality of characteristic parameters, a predetermined number of domains having been predefined in a space of the spatial distribution, the domains including fixed domains and one other domain, each domain defining a respective subspace;
calculating presumed information for each of said plurality of categories from part of the plurality of measured data vectors, the presumed information including a plurality of presumed centers-of-gravity;
determining a corrected center-of-gravity for each category based on the presumed center-of-gravity being used as an initial center-of-gravity, the step of determining a corrected center-of-gravity including initially dividing said particles into said plurality of categories based on the spatial distribution of the measured data vectors and then determining a membership value set for each measured data vector, each membership value set including a membership value corresponding to each domain and being a weight of membership of the measured data vector in the category corresponding to the domain; and
clustering said mixture of particles into said plurality of categories based on the spatial distribution of said measured data vectors and based on said corrected center-of-gravity and counting a number of particles belonging to each of the categories of particles based on the membership value sets, wherein the step of determining a membership value set for each measured data vector includes:
when a measured data vector is located in the subspace of said one other domain, calculating the membership value set for each measured data vector based on a distance defined between the measured data vector and the initial center-of-gravity of each category; and
when a measured data vector is located in the subspace of a particular domain of the fixed domains, assigning a first value to the membership value corresponding to the particular domain and assigning a second value to the membership value corresponding to another fixed domain different than the particular domain.
2. The method of claim 1, wherein the presumed information includes a presumed count and a presumed variance for each category, and the step of determining a membership value set for each measured data vector further includes assuming the presumed count and the presumed variance to be an initial count and an initial variance for each category, and the step of calculating the membership value set for each measured data vector includes:
calculating a plurality of non-Euclidean distances, each non-Euclidean distance corresponding to a category of the plurality of categories and being defined by a function of a distance between the measured data vector and the initial center-of-gravity of the category; and
assigning membership values to the membership value set based on at least one of the non-Euclidean distance, the initial count and the initial variance of the respective category.
3. In an analyzing device, a method of analyzing a mixture of particles belonging to a plurality of categories to count a number of particles belonging to each of the categories, each particle being individually characterized by a corresponding measured data vector, each measured data vector including a plurality of characteristic parameters, each characteristic parameter being representative of a physical characteristic of a corresponding particle, the method comprising steps of:
converting a plurality of measured data vectors representative of a specimen of particles into a spatial distribution of the plurality of measured data vectors based on spatial dimensions corresponding to the plurality of characteristic parameters, a predetermined number of domains having been predefined in a space of the spatial distribution, the domains including fixed domains and one other domain, each domain defining a respective subspace;
calculating presumed information for each of said plurality of categories from part of the plurality of measured data vectors, the presumed information including a plurality of presumed centers-of-gravity;
determining a corrected center-of-gravity for each category based on the presumed center-of-gravity being used as an initial center-of-gravity, the step of determining a corrected center-of-gravity including initially dividing said particles into said plurality of categories based on the spatial distribution of the measured dam vectors and then determining a membership value set for each measured dam vector, each membership value set including a membership value corresponding to each domain and being a weight of membership of the measured data vector in the category corresponding to the domain; and
clustering said mixture of particles into said plurality of categories based on the spatial distribution of said measured data vectors and based on said corrected center-of-gravity and counting a number of particles belonging to each of the categories of particles based on the membership value sets, wherein the step of determining a corrected center-of-gravity further includes:
calculating the corrected center-of-gravity for each category based on the membership value sets;
determining a difference between the corrected center-of-gravity and the initial center-of-gravity for each category; and
judging that the number of particles belonging to each of the categories of particles is not yet ascertained when the difference determined for any of the categories is greater than a predetermined difference threshold.
4. The method of claim 3, wherein when the number of particles belonging to each of the categories of particles is judged to be not yet ascertained, the step of determining a corrected center-of-gravity further includes steps of:
updating the initial center-of-gravity with the corrected center-of-gravity for each category; and
repeating the steps of determining a membership value set for each measured data vector, calculating a corrected center-of-gravity, and determining a difference.
5. The method of claim 4, wherein the step of repeating is successively repeated until an occurrence of an end condition, the end condition being one of:
when a predetermined number of cycles of the step of determining a difference has occurred; and
when the difference determined for each category is less than the predetermined difference threshold.
6. An analyzing device for analyzing a mixture of particles belonging to a plurality of categories to count a number of particles belonging to each of the categories, each particle being individually characterized by a corresponding measured data vector, each measured data vector including a plurality of characteristic parameters, each characteristic parameter being representative of a physical characteristic of a corresponding particle, the device comprising;
means for converting a plurality of measured data vectors representative of a specimen of particles into a spatial distribution of the plurality of measured data vectors based on spatial dimensions corresponding to the plurality of characteristic parameters, a predetermined number of domains having been predefined in a space of the spatial distribution, the domains including fixed domains and one other domain, each domain defining a respective subspace;
means for calculating presumed information for each of said plurality of categories from part of the plurality of measured data vectors, the presumed information including a plurality of presumed centers-of-gravity;
means for determining a corrected center-of-gravity for each category based on the presumed center-of-gravity being used as an initial center-of-gravity, the means for determining a corrected center-of-gravity including means for determining a membership value set for each measured data vector, each membership value set including a membership value corresponding to each domain and being a weight of membership of the measured data vector in the category corresponding to the domain; and
means for clustering said mixture of particles into said plurality of categories based on the spatial distribution of said measured data vectors and based on said corrected center-of-gravity and for counting a number of particles belonging to each of the categories of particles based on the membership value sets, wherein the means for determining a membership value set for each measured data vector includes:
means, operable when a measured data vector is located in the subspace of said one other domain, for calculating the membership value set for each measured data vector based on a distance defined between the measured data vector and the initial center-of-gravity of each category; and
means, operable when the measured data vector is located in the subspace of a particular domain of the fixed domains, for assigning a first value to the membership value corresponding to the particular domain and for assigning a second value to the membership value corresponding to another fixed domain different than the particular domain.
7. The device of claim 6, wherein the information includes a presumed count and a presumed variance for each category, the means for determining a membership value set for each measured data vector includes assuming the presumed count and the presumed variance to be an initial count and an initial variance for each category and the means for calculating the membership value set for each measured data vector includes:
means for calculating a plurality of non-Euclidean distances, each non-Euclidean distance corresponding to a category of the plurality of categories and being defined by a function of a distance between the measured data vector and the initial center-of-gravity of the category; and
means for assigning membership values to the membership value set based on at least one of the non-Euclidean distance, the initial count and the initial variance of the respective category.
8. An analyzing device for analyzing a mixture of particles belonging to a plurality of categories to count a number of particles belonging to each of the categories, each particle being individually characterized by a corresponding measured data vector, each measured data vector including a plurality of characteristic parameters, each characteristic parameter being representative of a physical characteristic of a corresponding particle, the device comprising:
means for converting a plurality of measured data vectors representative of a specimen particles into a spatial distribution of the plurality of measured data vectors based on spatial dimensions corresponding to the plurality of characteristic parameters, a predetermined number of domains having been predefined in a space of the spatial distribution, the domains including fixed domains and one other domain, each domain defining a restrictive subspace;
means for calculating presumed information for each of said plurality of categories from part of the plurality of measured data vectors, the presumed information including a plurality of presumed centers-of-gravity;
means for determining a corrected center-of-gravity for each category based on the presumed center-of-gravity being used as an initial center-of-gravity, the means for determining a corrected center-of-gravity including means for determining a membership value set for each measured data vector, each membership value set including a membership value corresponding to each domain and being a weight of membership of the measured data vector in the category corresponding to the domain; and
means for clustering said mixture of particles into said plurality of categories based on the spatial distribution of said measured data vectors and based on said corrected center-of-gravity and for counting a number of particles belonging to each of the categories of particles based on the membership value sets, wherein the means for determining a corrected center-of-gravity includes:
means for calculating the corrected center-of-gravity for each category based on the membership value sets;
means for determining a difference between the corrected center-of-gravity and the initial center-of-gravity for each category; and
means for judging that the number of particles belonging to each of the categories of particles is not yet ascertained when the difference determined for any of the categories is greater than a predetermined difference threshold.
9. The device of claim 8, wherein the means for determining a corrected center-of-gravity includes means, operable when the number of particles belonging to each of the categories of particles is judged to be not yet ascertained, for updating the initial center-of-gravity with the corrected center-of-gravity for each category, and for repeating an operation of the means for determining a membership value set for each measured data vector, the means for calculating the corrected center-of-gravity, and the means for determining a difference.
10. The device of claim 9, wherein an operation of the means for updating and repealing is successively repeated until an occurrence of an end condition, the end condition being one of:
when a predetermined number of operations of the means for updating and repeating has occurred; and
when the difference determined for each category is less than the predetermined difference threshold.
US08368053 1992-06-19 1995-01-03 Method and device for analyzing particles Expired - Lifetime US5703959A (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP18626592A JP3165247B2 (en) 1992-06-19 1992-06-19 Particle analyzing method and apparatus
JP4-186265 1992-06-19
US7687093 true 1993-06-15 1993-06-15
US08368053 US5703959A (en) 1992-06-19 1995-01-03 Method and device for analyzing particles

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US08368053 US5703959A (en) 1992-06-19 1995-01-03 Method and device for analyzing particles

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
US7687093 Continuation 1993-06-15 1993-06-15

Publications (1)

Publication Number Publication Date
US5703959A true US5703959A (en) 1997-12-30

Family

ID=16185263

Family Applications (1)

Application Number Title Priority Date Filing Date
US08368053 Expired - Lifetime US5703959A (en) 1992-06-19 1995-01-03 Method and device for analyzing particles

Country Status (5)

Country Link
US (1) US5703959A (en)
EP (1) EP0575091B1 (en)
JP (1) JP3165247B2 (en)
CA (1) CA2098785A1 (en)
DE (2) DE69327615T2 (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5933519A (en) * 1994-09-20 1999-08-03 Neo Path, Inc. Cytological slide scoring apparatus
US6038340A (en) * 1996-11-08 2000-03-14 Seiko Epson Corporation System and method for detecting the black and white points of a color image
WO2002092161A1 (en) * 2001-05-10 2002-11-21 Biophan, Llc Miniaturized particle analyzer
WO2003011386A1 (en) * 2001-07-30 2003-02-13 Biomed Solutions, Llc Flow cytometer
US20030186461A1 (en) * 2002-03-29 2003-10-02 Cyrano Bioscienes, Inc. Method and system for using a weighted response
US6631211B1 (en) * 1999-07-08 2003-10-07 Perkinelmer Las, Inc. Interactive system for analyzing scatter plots
US6633669B1 (en) * 1999-10-21 2003-10-14 3M Innovative Properties Company Autogrid analysis
US6662117B2 (en) * 2000-07-24 2003-12-09 Sysmex Corporation Particle analyzer and particle classifying method
US6678040B1 (en) * 1999-07-02 2004-01-13 Terumo Kabushiki Kaisha Apparatus for measuring number of cells
US20040126008A1 (en) * 2000-04-24 2004-07-01 Eric Chapoulaud Analyte recognition for urinalysis diagnostic system
US20050008228A1 (en) * 2003-05-23 2005-01-13 Eric Chapoulaud Fluid sample analysis using class weights
US6943358B1 (en) * 2001-03-21 2005-09-13 The United States Of America As Represented By The Secretary Of The Navy Method for developing a calibration algorithm for quantifying the hydrocarbon content of aqueous media
WO2006119482A2 (en) * 2005-05-04 2006-11-09 West Virginia University Research Corporation Method for data clustering and classification by a graph theory model -- network partition into high density subgraphs
US20070003135A1 (en) * 2004-02-27 2007-01-04 Mattausch Hans J Reference data optimization learning method and pattern recognition system
US20120155739A1 (en) * 2009-07-20 2012-06-21 Fraunhofer-Gesellschaft Zur Foerderung Der Angewandten Forschung E.V. Localization of a valid area of a blood smear

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE69429145D1 (en) * 1993-08-19 2002-01-03 Hitachi Ltd Classification and examination device of particles in a liquid
EP0660104A3 (en) * 1993-12-27 1995-08-09 Hitachi Ltd Urinary sediment examining apparatus.
JP3050046B2 (en) * 1994-07-18 2000-06-05 株式会社日立製作所 Particles automatic classification system
FR2733596B1 (en) * 1995-04-28 1997-07-18 Hycel Groupe Lisabio Method and particle identification device
FR2754346B1 (en) * 1996-10-07 1999-02-05 Hycel Groupe Lisabio Method for identification of groups of particles and identification device corresponding
US6979570B2 (en) 2001-07-26 2005-12-27 Sysmex Corporation Particle analyzer and particle analyzing method
GB0523475D0 (en) * 2004-11-19 2005-12-28 Hull And East Yorkshire Hospit Diagnostic apparatus
JP5168658B2 (en) * 2008-11-21 2013-03-21 国立大学法人高知大学 Blood cell analyzer, blood cell analyzing method, and computer program
WO2013037119A1 (en) * 2011-09-16 2013-03-21 长沙高新技术产业开发区爱威科技实业有限公司 Device and method for erythrocyte morphology analysis
CN106021852A (en) * 2016-05-06 2016-10-12 鼎泰生物科技(海南)有限公司 Density clustering algorithm based blood glucose data abnormal value computing method and device

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4097845A (en) * 1976-11-01 1978-06-27 Rush-Presbyterian-St. Luke's Medical Center Method of and an apparatus for automatic classification of red blood cells
US4661913A (en) * 1984-09-11 1987-04-28 Becton, Dickinson And Company Apparatus and method for the detection and classification of articles using flow cytometry techniques
US4791355A (en) * 1986-10-21 1988-12-13 Coulter Electronics Inc. Particle analyzer for measuring the resistance and reactance of a particle
US4965725A (en) * 1988-04-08 1990-10-23 Nueromedical Systems, Inc. Neural network based automated cytological specimen classification system and method
JPH03131756A (en) * 1989-10-18 1991-06-05 Hitachi Ltd Automatic classifying device for blood cell
EP0544385A2 (en) * 1991-11-27 1993-06-02 Toa Medical Electronics Company, Limited Method of counting particles
US5255346A (en) * 1989-12-28 1993-10-19 U S West Advanced Technologies, Inc. Method and apparatus for design of a vector quantizer
US5627040A (en) * 1991-08-28 1997-05-06 Becton Dickinson And Company Flow cytometric method for autoclustering cells

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4097845A (en) * 1976-11-01 1978-06-27 Rush-Presbyterian-St. Luke's Medical Center Method of and an apparatus for automatic classification of red blood cells
US4661913A (en) * 1984-09-11 1987-04-28 Becton, Dickinson And Company Apparatus and method for the detection and classification of articles using flow cytometry techniques
US4791355A (en) * 1986-10-21 1988-12-13 Coulter Electronics Inc. Particle analyzer for measuring the resistance and reactance of a particle
US4965725A (en) * 1988-04-08 1990-10-23 Nueromedical Systems, Inc. Neural network based automated cytological specimen classification system and method
US4965725B1 (en) * 1988-04-08 1996-05-07 Neuromedical Systems Inc Neural network based automated cytological specimen classification system and method
JPH03131756A (en) * 1989-10-18 1991-06-05 Hitachi Ltd Automatic classifying device for blood cell
US5255346A (en) * 1989-12-28 1993-10-19 U S West Advanced Technologies, Inc. Method and apparatus for design of a vector quantizer
US5627040A (en) * 1991-08-28 1997-05-06 Becton Dickinson And Company Flow cytometric method for autoclustering cells
EP0544385A2 (en) * 1991-11-27 1993-06-02 Toa Medical Electronics Company, Limited Method of counting particles
US5555196A (en) * 1991-11-27 1996-09-10 Toa Medical Electronics Co., Ltd. Method of counting particles using degree of membership in clustering data points into subgroups
US5555198A (en) * 1991-11-27 1996-09-10 Toa Medical Electronics Co., Ltd. Apparatus for counting particles

Non-Patent Citations (16)

* Cited by examiner, † Cited by third party
Title
Gulcur et al. "Identification of Different Types of Leucocytes in Dried Blood Smears Using Neural Networks," Proc. 1992 Int. Biomedical Eng. Days, Aug. 1992, pp. 203-206.
Gulcur et al. Identification of Different Types of Leucocytes in Dried Blood Smears Using Neural Networks, Proc. 1992 Int. Biomedical Eng. Days, Aug. 1992, pp. 203 206. *
J. Bezdek et al., "Fuzzy Kohonen Clustering Networks", IEEE International Conference on Fuzzy Systems, Mar. 8-12, 1992, San Diego, CA, USA, pp. 1035-1043.
J. Bezdek et al., Fuzzy Kohonen Clustering Networks , IEEE International Conference on Fuzzy Systems, Mar. 8 12, 1992, San Diego, CA, USA, pp. 1035 1043. *
J. Weber et al., "Guided Data Reduction for Flow Cytometry", Images of the Twenty-First Century, Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Seattle, WA, USA, vol. 11:1989; pp. 1783-1784.
J. Weber et al., Guided Data Reduction for Flow Cytometry , Images of the Twenty First Century, Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Seattle, WA, USA, vol. 11:1989; pp. 1783 1784. *
Johnson et al. "Multidimensional Self Organisation." 1990 IEEE Int. Workshop on Cellular Neural Networks and Their Appl., Dec. 1990, pp. 254-263.
Johnson et al. Multidimensional Self Organisation. 1990 IEEE Int. Workshop on Cellular Neural Networks and Their Appl., Dec. 1990, pp. 254 263. *
Kelley et al. "An Adaptive Algorithm For Modifying Hyperellipsoidal Decision Surfaces", 7-11 Jun., 1992 pp. IV-196 to IV-201, IJCNN International Joint Conference on Neural Networks.
Kelley et al. An Adaptive Algorithm For Modifying Hyperellipsoidal Decision Surfaces , 7 11 Jun., 1992 pp. IV 196 to IV 201, IJCNN International Joint Conference on Neural Networks. *
L. Fu et al, "A Hybrid System Approach to Multivariate Analysis of Flow Cytometry Data", Proceedings of the Fifth Annual IEEE Symposium on Computer-Based Medical Systems, ISBN 0-8186-2742-5, Jun. 1992, pp. 315-324.
L. Fu et al, A Hybrid System Approach to Multivariate Analysis of Flow Cytometry Data , Proceedings of the Fifth Annual IEEE Symposium on Computer Based Medical Systems, ISBN 0 8186 2742 5, Jun. 1992, pp. 315 324. *
S G Kong et al., Differential Competitive Learning For Phoneme Recognition , Neural Networks For Signal Processing, pp. 1 32. *
S. Newton et al., "Adaptive Fuzzy Leader Clustering of Complex Data Sets in Pattern Recognition", IEEE Transactions on Neural Networks, vol. 3, No. 5, Sep. 1992, pp. 794-800.
S. Newton et al., Adaptive Fuzzy Leader Clustering of Complex Data Sets in Pattern Recognition , IEEE Transactions on Neural Networks, vol. 3, No. 5, Sep. 1992, pp. 794 800. *
S-G Kong et al., "Differential Competitive Learning For Phoneme Recognition", Neural Networks For Signal Processing, pp. 1-32.

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5933519A (en) * 1994-09-20 1999-08-03 Neo Path, Inc. Cytological slide scoring apparatus
US6038340A (en) * 1996-11-08 2000-03-14 Seiko Epson Corporation System and method for detecting the black and white points of a color image
US6678040B1 (en) * 1999-07-02 2004-01-13 Terumo Kabushiki Kaisha Apparatus for measuring number of cells
US6631211B1 (en) * 1999-07-08 2003-10-07 Perkinelmer Las, Inc. Interactive system for analyzing scatter plots
US6714675B1 (en) * 1999-10-21 2004-03-30 3M Innovative Properties Company Autogrid analysis
US6633669B1 (en) * 1999-10-21 2003-10-14 3M Innovative Properties Company Autogrid analysis
US20040126008A1 (en) * 2000-04-24 2004-07-01 Eric Chapoulaud Analyte recognition for urinalysis diagnostic system
US7236623B2 (en) * 2000-04-24 2007-06-26 International Remote Imaging Systems, Inc. Analyte recognition for urinalysis diagnostic system
US6662117B2 (en) * 2000-07-24 2003-12-09 Sysmex Corporation Particle analyzer and particle classifying method
US6943358B1 (en) * 2001-03-21 2005-09-13 The United States Of America As Represented By The Secretary Of The Navy Method for developing a calibration algorithm for quantifying the hydrocarbon content of aqueous media
US6793642B2 (en) 2001-05-07 2004-09-21 Biomed Solutions, Llc Flow cytometer
WO2002092161A1 (en) * 2001-05-10 2002-11-21 Biophan, Llc Miniaturized particle analyzer
WO2003011386A1 (en) * 2001-07-30 2003-02-13 Biomed Solutions, Llc Flow cytometer
US20030186461A1 (en) * 2002-03-29 2003-10-02 Cyrano Bioscienes, Inc. Method and system for using a weighted response
US7129095B2 (en) * 2002-03-29 2006-10-31 Smiths Detection Inc. Method and system for using a weighted response
US7324694B2 (en) * 2003-05-23 2008-01-29 International Remote Imaging Systems, Inc. Fluid sample analysis using class weights
US20050008228A1 (en) * 2003-05-23 2005-01-13 Eric Chapoulaud Fluid sample analysis using class weights
US20070003135A1 (en) * 2004-02-27 2007-01-04 Mattausch Hans J Reference data optimization learning method and pattern recognition system
US7881525B2 (en) * 2004-02-27 2011-02-01 Hiroshima University Reference data optimization learning method and pattern recognition system
US20060274062A1 (en) * 2005-05-04 2006-12-07 Cun-Quan Zhang Method for data clustering and classification by a graph theory model - network partition into high density subgraphs
WO2006119482A2 (en) * 2005-05-04 2006-11-09 West Virginia University Research Corporation Method for data clustering and classification by a graph theory model -- network partition into high density subgraphs
WO2006119482A3 (en) * 2005-05-04 2009-04-16 Univ West Virginia Method for data clustering and classification by a graph theory model -- network partition into high density subgraphs
US7523117B2 (en) * 2005-05-04 2009-04-21 West Virginia University Research Corporation Method for data clustering and classification by a graph theory model—network partition into high density subgraphs
US20120155739A1 (en) * 2009-07-20 2012-06-21 Fraunhofer-Gesellschaft Zur Foerderung Der Angewandten Forschung E.V. Localization of a valid area of a blood smear
US8538122B2 (en) * 2009-07-20 2013-09-17 Fraunhofer-Gesellschaft Zur Foerderung Der Angewandten Forschung E.V. Localization of a valid area of a blood smear

Also Published As

Publication number Publication date Type
JPH063252A (en) 1994-01-11 application
DE69327615T2 (en) 2000-08-03 grant
DE69327615D1 (en) 2000-02-24 grant
CA2098785A1 (en) 1993-12-20 application
JP3165247B2 (en) 2001-05-14 grant
EP0575091A2 (en) 1993-12-22 application
EP0575091A3 (en) 1994-08-24 application
EP0575091B1 (en) 2000-01-19 grant

Similar Documents

Publication Publication Date Title
Raudys et al. Small sample size effects in statistical pattern recognition: Recommendations for practitioners
Zorriassatine et al. A review of neural networks for statistical process control
Fletcher et al. Forecasting with neural networks: an application using bankruptcy data
Huntsherger et al. Iterative fuzzy image segmentation
Odewahn et al. Automated star/galaxy discrimination with neural networks
Ruck et al. The multilayer perceptron as an approximation to a Bayes optimal discriminant function
US5299284A (en) Pattern classification using linear programming
US5799101A (en) Method and apparatus for highly efficient computer aided screening
Specht Probabilistic neural networks and the polynomial adaline as complementary techniques for classification
US5155801A (en) Clustered neural networks
Tam Neural network models and the prediction of bank bankruptcy
Chiu Extracting fuzzy rules from data for function approximation and pattern classification
Windham Cluster validity for fuzzy clustering algorithms
Aleksander et al. WISARD⊙ a radical step forward in image recognition
Gish et al. Segregation of speakers for speech recognition and speaker identification
Parasuraman et al. Fuzzy signal detection theory: Basic postulates and formulas for analyzing human and machine performance
Nakano Application of neural networks to the color grading of apples
US20020159630A1 (en) Automated detection of pornographic images
Kranz Diagnosis of partial discharge signals using neural networks and minimum distance classification
Spector What to do with significant multivariate effects in multivariate analyses of variance.
Yeung et al. Parzen-window network intrusion detectors
US5329596A (en) Automatic clustering method
Rudasi et al. Text-independent talker identification with neural networks
Levin et al. Accelerated learning in layered neural networks
US6556951B1 (en) System and method for intelligent quality control of a process

Legal Events

Date Code Title Description
FPAY Fee payment

Year of fee payment: 4

FPAY Fee payment

Year of fee payment: 8

FPAY Fee payment

Year of fee payment: 12